Mercurial > libavcodec.hg
annotate jfdctint.c @ 2822:fdedaa2e6da4 libavcodec
DV video encoder/decoder doesn't set keyframe and picture type properties patch by (Edward Hervey:bilboed,gmail com)
author | michael |
---|---|
date | Thu, 11 Aug 2005 01:28:04 +0000 |
parents | eb26d190cf5a |
children | ef2149182f1c |
rev | line source |
---|---|
625
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
1 /* |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
2 * jfdctint.c |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
3 * |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
4 * Copyright (C) 1991-1996, Thomas G. Lane. |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
5 * This file is part of the Independent JPEG Group's software. |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
6 * For conditions of distribution and use, see the accompanying README file. |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
7 * |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
8 * This file contains a slow-but-accurate integer implementation of the |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
9 * forward DCT (Discrete Cosine Transform). |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
10 * |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
11 * A 2-D DCT can be done by 1-D DCT on each row followed by 1-D DCT |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
12 * on each column. Direct algorithms are also available, but they are |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
13 * much more complex and seem not to be any faster when reduced to code. |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
14 * |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
15 * This implementation is based on an algorithm described in |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
16 * C. Loeffler, A. Ligtenberg and G. Moschytz, "Practical Fast 1-D DCT |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
17 * Algorithms with 11 Multiplications", Proc. Int'l. Conf. on Acoustics, |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
18 * Speech, and Signal Processing 1989 (ICASSP '89), pp. 988-991. |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
19 * The primary algorithm described there uses 11 multiplies and 29 adds. |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
20 * We use their alternate method with 12 multiplies and 32 adds. |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
21 * The advantage of this method is that no data path contains more than one |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
22 * multiplication; this allows a very simple and accurate implementation in |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
23 * scaled fixed-point arithmetic, with a minimal number of shifts. |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
24 */ |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
25 |
1106 | 26 /** |
27 * @file jfdctint.c | |
28 * Independent JPEG Group's slow & accurate dct. | |
29 */ | |
30 | |
625
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
31 #include <stdlib.h> |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
32 #include <stdio.h> |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
33 #include "common.h" |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
34 #include "dsputil.h" |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
35 |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
36 #define SHIFT_TEMPS |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
37 #define DCTSIZE 8 |
637 | 38 #define BITS_IN_JSAMPLE 8 |
625
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
39 #define GLOBAL(x) x |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
40 #define RIGHT_SHIFT(x, n) ((x) >> (n)) |
637 | 41 #define MULTIPLY16C16(var,const) ((var)*(const)) |
625
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
42 |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
43 #if 1 //def USE_ACCURATE_ROUNDING |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
44 #define DESCALE(x,n) RIGHT_SHIFT((x) + (1 << ((n) - 1)), n) |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
45 #else |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
46 #define DESCALE(x,n) RIGHT_SHIFT(x, n) |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
47 #endif |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
48 |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
49 |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
50 /* |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
51 * This module is specialized to the case DCTSIZE = 8. |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
52 */ |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
53 |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
54 #if DCTSIZE != 8 |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
55 Sorry, this code only copes with 8x8 DCTs. /* deliberate syntax err */ |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
56 #endif |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
57 |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
58 |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
59 /* |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
60 * The poop on this scaling stuff is as follows: |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
61 * |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
62 * Each 1-D DCT step produces outputs which are a factor of sqrt(N) |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
63 * larger than the true DCT outputs. The final outputs are therefore |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
64 * a factor of N larger than desired; since N=8 this can be cured by |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
65 * a simple right shift at the end of the algorithm. The advantage of |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
66 * this arrangement is that we save two multiplications per 1-D DCT, |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
67 * because the y0 and y4 outputs need not be divided by sqrt(N). |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
68 * In the IJG code, this factor of 8 is removed by the quantization step |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
69 * (in jcdctmgr.c), NOT in this module. |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
70 * |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
71 * We have to do addition and subtraction of the integer inputs, which |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
72 * is no problem, and multiplication by fractional constants, which is |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
73 * a problem to do in integer arithmetic. We multiply all the constants |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
74 * by CONST_SCALE and convert them to integer constants (thus retaining |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
75 * CONST_BITS bits of precision in the constants). After doing a |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
76 * multiplication we have to divide the product by CONST_SCALE, with proper |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
77 * rounding, to produce the correct output. This division can be done |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
78 * cheaply as a right shift of CONST_BITS bits. We postpone shifting |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
79 * as long as possible so that partial sums can be added together with |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
80 * full fractional precision. |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
81 * |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
82 * The outputs of the first pass are scaled up by PASS1_BITS bits so that |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
83 * they are represented to better-than-integral precision. These outputs |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
84 * require BITS_IN_JSAMPLE + PASS1_BITS + 3 bits; this fits in a 16-bit word |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
85 * with the recommended scaling. (For 12-bit sample data, the intermediate |
1064 | 86 * array is int32_t anyway.) |
625
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
87 * |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
88 * To avoid overflow of the 32-bit intermediate results in pass 2, we must |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
89 * have BITS_IN_JSAMPLE + CONST_BITS + PASS1_BITS <= 26. Error analysis |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
90 * shows that the values given below are the most effective. |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
91 */ |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
92 |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
93 #if BITS_IN_JSAMPLE == 8 |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
94 #define CONST_BITS 13 |
637 | 95 #define PASS1_BITS 4 /* set this to 2 if 16x16 multiplies are faster */ |
625
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
96 #else |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
97 #define CONST_BITS 13 |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
98 #define PASS1_BITS 1 /* lose a little precision to avoid overflow */ |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
99 #endif |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
100 |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
101 /* Some C compilers fail to reduce "FIX(constant)" at compile time, thus |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
102 * causing a lot of useless floating-point operations at run time. |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
103 * To get around this we use the following pre-calculated constants. |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
104 * If you change CONST_BITS you may want to add appropriate values. |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
105 * (With a reasonable C compiler, you can just rely on the FIX() macro...) |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
106 */ |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
107 |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
108 #if CONST_BITS == 13 |
1064 | 109 #define FIX_0_298631336 ((int32_t) 2446) /* FIX(0.298631336) */ |
110 #define FIX_0_390180644 ((int32_t) 3196) /* FIX(0.390180644) */ | |
111 #define FIX_0_541196100 ((int32_t) 4433) /* FIX(0.541196100) */ | |
112 #define FIX_0_765366865 ((int32_t) 6270) /* FIX(0.765366865) */ | |
113 #define FIX_0_899976223 ((int32_t) 7373) /* FIX(0.899976223) */ | |
114 #define FIX_1_175875602 ((int32_t) 9633) /* FIX(1.175875602) */ | |
115 #define FIX_1_501321110 ((int32_t) 12299) /* FIX(1.501321110) */ | |
116 #define FIX_1_847759065 ((int32_t) 15137) /* FIX(1.847759065) */ | |
117 #define FIX_1_961570560 ((int32_t) 16069) /* FIX(1.961570560) */ | |
118 #define FIX_2_053119869 ((int32_t) 16819) /* FIX(2.053119869) */ | |
119 #define FIX_2_562915447 ((int32_t) 20995) /* FIX(2.562915447) */ | |
120 #define FIX_3_072711026 ((int32_t) 25172) /* FIX(3.072711026) */ | |
625
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
121 #else |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
122 #define FIX_0_298631336 FIX(0.298631336) |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
123 #define FIX_0_390180644 FIX(0.390180644) |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
124 #define FIX_0_541196100 FIX(0.541196100) |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
125 #define FIX_0_765366865 FIX(0.765366865) |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
126 #define FIX_0_899976223 FIX(0.899976223) |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
127 #define FIX_1_175875602 FIX(1.175875602) |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
128 #define FIX_1_501321110 FIX(1.501321110) |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
129 #define FIX_1_847759065 FIX(1.847759065) |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
130 #define FIX_1_961570560 FIX(1.961570560) |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
131 #define FIX_2_053119869 FIX(2.053119869) |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
132 #define FIX_2_562915447 FIX(2.562915447) |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
133 #define FIX_3_072711026 FIX(3.072711026) |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
134 #endif |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
135 |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
136 |
1064 | 137 /* Multiply an int32_t variable by an int32_t constant to yield an int32_t result. |
625
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
138 * For 8-bit samples with the recommended scaling, all the variable |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
139 * and constant values involved are no more than 16 bits wide, so a |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
140 * 16x16->32 bit multiply can be used instead of a full 32x32 multiply. |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
141 * For 12-bit samples, a full 32-bit multiplication will be needed. |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
142 */ |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
143 |
637 | 144 #if BITS_IN_JSAMPLE == 8 && CONST_BITS<=13 && PASS1_BITS<=2 |
625
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
145 #define MULTIPLY(var,const) MULTIPLY16C16(var,const) |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
146 #else |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
147 #define MULTIPLY(var,const) ((var) * (const)) |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
148 #endif |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
149 |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
150 |
1589 | 151 static always_inline void row_fdct(DCTELEM * data){ |
152 int_fast32_t tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7; | |
153 int_fast32_t tmp10, tmp11, tmp12, tmp13; | |
154 int_fast32_t z1, z2, z3, z4, z5; | |
625
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
155 DCTELEM *dataptr; |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
156 int ctr; |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
157 SHIFT_TEMPS |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
158 |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
159 /* Pass 1: process rows. */ |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
160 /* Note results are scaled up by sqrt(8) compared to a true DCT; */ |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
161 /* furthermore, we scale the results by 2**PASS1_BITS. */ |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
162 |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
163 dataptr = data; |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
164 for (ctr = DCTSIZE-1; ctr >= 0; ctr--) { |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
165 tmp0 = dataptr[0] + dataptr[7]; |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
166 tmp7 = dataptr[0] - dataptr[7]; |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
167 tmp1 = dataptr[1] + dataptr[6]; |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
168 tmp6 = dataptr[1] - dataptr[6]; |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
169 tmp2 = dataptr[2] + dataptr[5]; |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
170 tmp5 = dataptr[2] - dataptr[5]; |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
171 tmp3 = dataptr[3] + dataptr[4]; |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
172 tmp4 = dataptr[3] - dataptr[4]; |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
173 |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
174 /* Even part per LL&M figure 1 --- note that published figure is faulty; |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
175 * rotator "sqrt(2)*c1" should be "sqrt(2)*c6". |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
176 */ |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
177 |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
178 tmp10 = tmp0 + tmp3; |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
179 tmp13 = tmp0 - tmp3; |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
180 tmp11 = tmp1 + tmp2; |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
181 tmp12 = tmp1 - tmp2; |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
182 |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
183 dataptr[0] = (DCTELEM) ((tmp10 + tmp11) << PASS1_BITS); |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
184 dataptr[4] = (DCTELEM) ((tmp10 - tmp11) << PASS1_BITS); |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
185 |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
186 z1 = MULTIPLY(tmp12 + tmp13, FIX_0_541196100); |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
187 dataptr[2] = (DCTELEM) DESCALE(z1 + MULTIPLY(tmp13, FIX_0_765366865), |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
188 CONST_BITS-PASS1_BITS); |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
189 dataptr[6] = (DCTELEM) DESCALE(z1 + MULTIPLY(tmp12, - FIX_1_847759065), |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
190 CONST_BITS-PASS1_BITS); |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
191 |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
192 /* Odd part per figure 8 --- note paper omits factor of sqrt(2). |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
193 * cK represents cos(K*pi/16). |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
194 * i0..i3 in the paper are tmp4..tmp7 here. |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
195 */ |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
196 |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
197 z1 = tmp4 + tmp7; |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
198 z2 = tmp5 + tmp6; |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
199 z3 = tmp4 + tmp6; |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
200 z4 = tmp5 + tmp7; |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
201 z5 = MULTIPLY(z3 + z4, FIX_1_175875602); /* sqrt(2) * c3 */ |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
202 |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
203 tmp4 = MULTIPLY(tmp4, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */ |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
204 tmp5 = MULTIPLY(tmp5, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */ |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
205 tmp6 = MULTIPLY(tmp6, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */ |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
206 tmp7 = MULTIPLY(tmp7, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */ |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
207 z1 = MULTIPLY(z1, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */ |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
208 z2 = MULTIPLY(z2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */ |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
209 z3 = MULTIPLY(z3, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */ |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
210 z4 = MULTIPLY(z4, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */ |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
211 |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
212 z3 += z5; |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
213 z4 += z5; |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
214 |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
215 dataptr[7] = (DCTELEM) DESCALE(tmp4 + z1 + z3, CONST_BITS-PASS1_BITS); |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
216 dataptr[5] = (DCTELEM) DESCALE(tmp5 + z2 + z4, CONST_BITS-PASS1_BITS); |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
217 dataptr[3] = (DCTELEM) DESCALE(tmp6 + z2 + z3, CONST_BITS-PASS1_BITS); |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
218 dataptr[1] = (DCTELEM) DESCALE(tmp7 + z1 + z4, CONST_BITS-PASS1_BITS); |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
219 |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
220 dataptr += DCTSIZE; /* advance pointer to next row */ |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
221 } |
1589 | 222 } |
223 | |
224 /* | |
225 * Perform the forward DCT on one block of samples. | |
226 */ | |
227 | |
228 GLOBAL(void) | |
229 ff_jpeg_fdct_islow (DCTELEM * data) | |
230 { | |
231 int_fast32_t tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7; | |
232 int_fast32_t tmp10, tmp11, tmp12, tmp13; | |
233 int_fast32_t z1, z2, z3, z4, z5; | |
234 DCTELEM *dataptr; | |
235 int ctr; | |
236 SHIFT_TEMPS | |
237 | |
238 row_fdct(data); | |
625
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
239 |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
240 /* Pass 2: process columns. |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
241 * We remove the PASS1_BITS scaling, but leave the results scaled up |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
242 * by an overall factor of 8. |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
243 */ |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
244 |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
245 dataptr = data; |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
246 for (ctr = DCTSIZE-1; ctr >= 0; ctr--) { |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
247 tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*7]; |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
248 tmp7 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*7]; |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
249 tmp1 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*6]; |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
250 tmp6 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*6]; |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
251 tmp2 = dataptr[DCTSIZE*2] + dataptr[DCTSIZE*5]; |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
252 tmp5 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*5]; |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
253 tmp3 = dataptr[DCTSIZE*3] + dataptr[DCTSIZE*4]; |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
254 tmp4 = dataptr[DCTSIZE*3] - dataptr[DCTSIZE*4]; |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
255 |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
256 /* Even part per LL&M figure 1 --- note that published figure is faulty; |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
257 * rotator "sqrt(2)*c1" should be "sqrt(2)*c6". |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
258 */ |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
259 |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
260 tmp10 = tmp0 + tmp3; |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
261 tmp13 = tmp0 - tmp3; |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
262 tmp11 = tmp1 + tmp2; |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
263 tmp12 = tmp1 - tmp2; |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
264 |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
265 dataptr[DCTSIZE*0] = (DCTELEM) DESCALE(tmp10 + tmp11, PASS1_BITS); |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
266 dataptr[DCTSIZE*4] = (DCTELEM) DESCALE(tmp10 - tmp11, PASS1_BITS); |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
267 |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
268 z1 = MULTIPLY(tmp12 + tmp13, FIX_0_541196100); |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
269 dataptr[DCTSIZE*2] = (DCTELEM) DESCALE(z1 + MULTIPLY(tmp13, FIX_0_765366865), |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
270 CONST_BITS+PASS1_BITS); |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
271 dataptr[DCTSIZE*6] = (DCTELEM) DESCALE(z1 + MULTIPLY(tmp12, - FIX_1_847759065), |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
272 CONST_BITS+PASS1_BITS); |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
273 |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
274 /* Odd part per figure 8 --- note paper omits factor of sqrt(2). |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
275 * cK represents cos(K*pi/16). |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
276 * i0..i3 in the paper are tmp4..tmp7 here. |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
277 */ |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
278 |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
279 z1 = tmp4 + tmp7; |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
280 z2 = tmp5 + tmp6; |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
281 z3 = tmp4 + tmp6; |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
282 z4 = tmp5 + tmp7; |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
283 z5 = MULTIPLY(z3 + z4, FIX_1_175875602); /* sqrt(2) * c3 */ |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
284 |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
285 tmp4 = MULTIPLY(tmp4, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */ |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
286 tmp5 = MULTIPLY(tmp5, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */ |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
287 tmp6 = MULTIPLY(tmp6, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */ |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
288 tmp7 = MULTIPLY(tmp7, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */ |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
289 z1 = MULTIPLY(z1, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */ |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
290 z2 = MULTIPLY(z2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */ |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
291 z3 = MULTIPLY(z3, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */ |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
292 z4 = MULTIPLY(z4, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */ |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
293 |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
294 z3 += z5; |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
295 z4 += z5; |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
296 |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
297 dataptr[DCTSIZE*7] = (DCTELEM) DESCALE(tmp4 + z1 + z3, |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
298 CONST_BITS+PASS1_BITS); |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
299 dataptr[DCTSIZE*5] = (DCTELEM) DESCALE(tmp5 + z2 + z4, |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
300 CONST_BITS+PASS1_BITS); |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
301 dataptr[DCTSIZE*3] = (DCTELEM) DESCALE(tmp6 + z2 + z3, |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
302 CONST_BITS+PASS1_BITS); |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
303 dataptr[DCTSIZE*1] = (DCTELEM) DESCALE(tmp7 + z1 + z4, |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
304 CONST_BITS+PASS1_BITS); |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
305 |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
306 dataptr++; /* advance pointer to next column */ |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
307 } |
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
diff
changeset
|
308 } |
1567 | 309 |
310 /* | |
311 * The secret of DCT2-4-8 is really simple -- you do the usual 1-DCT | |
312 * on the rows and then, instead of doing even and odd, part on the colums | |
313 * you do even part two times. | |
314 */ | |
315 GLOBAL(void) | |
316 ff_fdct248_islow (DCTELEM * data) | |
317 { | |
1589 | 318 int_fast32_t tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7; |
319 int_fast32_t tmp10, tmp11, tmp12, tmp13; | |
320 int_fast32_t z1; | |
1567 | 321 DCTELEM *dataptr; |
322 int ctr; | |
323 SHIFT_TEMPS | |
324 | |
1589 | 325 row_fdct(data); |
1567 | 326 |
327 /* Pass 2: process columns. | |
328 * We remove the PASS1_BITS scaling, but leave the results scaled up | |
329 * by an overall factor of 8. | |
330 */ | |
331 | |
332 dataptr = data; | |
333 for (ctr = DCTSIZE-1; ctr >= 0; ctr--) { | |
334 tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*1]; | |
335 tmp1 = dataptr[DCTSIZE*2] + dataptr[DCTSIZE*3]; | |
336 tmp2 = dataptr[DCTSIZE*4] + dataptr[DCTSIZE*5]; | |
337 tmp3 = dataptr[DCTSIZE*6] + dataptr[DCTSIZE*7]; | |
338 tmp4 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*1]; | |
339 tmp5 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*3]; | |
340 tmp6 = dataptr[DCTSIZE*4] - dataptr[DCTSIZE*5]; | |
341 tmp7 = dataptr[DCTSIZE*6] - dataptr[DCTSIZE*7]; | |
342 | |
343 tmp10 = tmp0 + tmp3; | |
344 tmp11 = tmp1 + tmp2; | |
345 tmp12 = tmp1 - tmp2; | |
346 tmp13 = tmp0 - tmp3; | |
347 | |
348 dataptr[DCTSIZE*0] = (DCTELEM) DESCALE(tmp10 + tmp11, PASS1_BITS); | |
349 dataptr[DCTSIZE*4] = (DCTELEM) DESCALE(tmp10 - tmp11, PASS1_BITS); | |
350 | |
351 z1 = MULTIPLY(tmp12 + tmp13, FIX_0_541196100); | |
352 dataptr[DCTSIZE*2] = (DCTELEM) DESCALE(z1 + MULTIPLY(tmp13, FIX_0_765366865), | |
353 CONST_BITS+PASS1_BITS); | |
354 dataptr[DCTSIZE*6] = (DCTELEM) DESCALE(z1 + MULTIPLY(tmp12, - FIX_1_847759065), | |
355 CONST_BITS+PASS1_BITS); | |
356 | |
357 tmp10 = tmp4 + tmp7; | |
358 tmp11 = tmp5 + tmp6; | |
359 tmp12 = tmp5 - tmp6; | |
360 tmp13 = tmp4 - tmp7; | |
361 | |
362 dataptr[DCTSIZE*1] = (DCTELEM) DESCALE(tmp10 + tmp11, PASS1_BITS); | |
363 dataptr[DCTSIZE*5] = (DCTELEM) DESCALE(tmp10 - tmp11, PASS1_BITS); | |
364 | |
365 z1 = MULTIPLY(tmp12 + tmp13, FIX_0_541196100); | |
366 dataptr[DCTSIZE*3] = (DCTELEM) DESCALE(z1 + MULTIPLY(tmp13, FIX_0_765366865), | |
367 CONST_BITS+PASS1_BITS); | |
368 dataptr[DCTSIZE*7] = (DCTELEM) DESCALE(z1 + MULTIPLY(tmp12, - FIX_1_847759065), | |
369 CONST_BITS+PASS1_BITS); | |
370 | |
371 dataptr++; /* advance pointer to next column */ | |
372 } | |
373 } |